Rename PaddedSize to AllocSize, in the hope that this
will make it more obvious what it represents, and stop
it being confused with the StoreSize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71349 91177308-0d34-0410-b5e6-96231b3b80d8

Use 16 bit arithmetic while retrieving the address of callee's frame during indirect function calls, and set pclath before every call to retrieve the frame address.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71323 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out code that optimize loop terminating condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71305 91177308-0d34-0410-b5e6-96231b3b80d8

Don't attempt to handle unsized types in ScalarEvolution's GEP analyzer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71302 91177308-0d34-0410-b5e6-96231b3b80d8

Enable loop bb placement optimization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71291 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bogus overflow checks by replacing them with actual
overflow checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71284 91177308-0d34-0410-b5e6-96231b3b80d8

PPC::B and PPC::BCC's target operand may be an immediate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71282 91177308-0d34-0410-b5e6-96231b3b80d8

In non-pic builds, we can use -mdynamic-no-pic for a little more speed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71281 91177308-0d34-0410-b5e6-96231b3b80d8

Use .td for tablegen files, not .ll.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71277 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid warning in release-asserts build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71275 91177308-0d34-0410-b5e6-96231b3b80d8

Mirror how Fast ISel determines if a region.end intrinsic is the end of an
inlined function or the end of a function. Before, this was never executing the
"inlined" version of the Record method.

This will become important once the inlined Dwarf writer patch lands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71268 91177308-0d34-0410-b5e6-96231b3b80d8

Fold trunc casts into add-recurrence expressions, allowing the
add-recurrence to be exposed. Add a new SCEV folding rule to
help simplify expressions in the presence of these extra truncs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71264 91177308-0d34-0410-b5e6-96231b3b80d8

Compute the offsets of the compile units. We need this so that when we emit a
concrete instance of an inlined function, we can get the actual address of the
abstract instance inside of the compile unit.

This isn't currently used, but will be by a future check-in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71263 91177308-0d34-0410-b5e6-96231b3b80d8

Fix another bug in r71252. This code supports GetElementPtr
constant exprs as well as instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71262 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71260 91177308-0d34-0410-b5e6-96231b3b80d8

Add memoization for getSCEVAtScope results for instructions
which are not analyzed with SCEV techniques, which can require
brute-forcing through a large number of instructions. This
fixes a massive compile-time issue on 400.perlbench (in
particular, the loop in MD5Transform).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71259 91177308-0d34-0410-b5e6-96231b3b80d8

Make the SCEV* form of getSCEVAtScope public, to allow ScalarEvolution
clients to use it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71258 91177308-0d34-0410-b5e6-96231b3b80d8

Minor clean ups. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71256 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an error from r71252.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71255 91177308-0d34-0410-b5e6-96231b3b80d8

Print out nicer dump info for DIDescriptor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71253 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out the code for creating SCEVs for GEPs into a
separate function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71252 91177308-0d34-0410-b5e6-96231b3b80d8

Implement several new SCEV folding rules for UDiv SCEVs.
This fixes an old FIXME, and is needed by some upcoming changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71247 91177308-0d34-0410-b5e6-96231b3b80d8

Add a getExitBlock utility function to LoopInfo which returns the
exit block of a loop, if there is exactly one, similar to
getExitingBlock.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71245 91177308-0d34-0410-b5e6-96231b3b80d8

Don't align loop header unless the loop back edge is below the header.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71242 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out cycle-finder code and make it generic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71241 91177308-0d34-0410-b5e6-96231b3b80d8

Do not emit bit tests if target does not support natively left shift

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71240 91177308-0d34-0410-b5e6-96231b3b80d8

Allow 8 bit select in custom inserter

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71239 91177308-0d34-0410-b5e6-96231b3b80d8

Properly expand libcalls for urem / srem. Also make code more straightforward.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71238 91177308-0d34-0410-b5e6-96231b3b80d8

Typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71237 91177308-0d34-0410-b5e6-96231b3b80d8

Expand UREM / SREM into libcalls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71236 91177308-0d34-0410-b5e6-96231b3b80d8

Add 8 bit select

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71235 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR4152: asm constraint validation happens before dag combine, so we
need to work a bit to combine things like (x+c1+c2) into x+c3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71232 91177308-0d34-0410-b5e6-96231b3b80d8

Add some examples to show the difference between
the various notions of type size used in LLVM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71230 91177308-0d34-0410-b5e6-96231b3b80d8

Change 'make install' to install tblgen, for better support of out-of-tree targets,
patch by Mikael Lepistö!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71226 91177308-0d34-0410-b5e6-96231b3b80d8

fix RewriteStoreUserOfWholeAlloca to use the correct type size
method, fixing a crash on PR4146.  While the store will
ultimately overwrite the "padded size" number of bits in memory,
the stored value may be a subset of this size.  This function
only wants to handle the case where all bits are stored.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71224 91177308-0d34-0410-b5e6-96231b3b80d8

Reverse branch condition only when there is a conditional branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71214 91177308-0d34-0410-b5e6-96231b3b80d8

Add explicit braces to disambiguate nested if/else. Removes a warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71211 91177308-0d34-0410-b5e6-96231b3b80d8

This transform requires valid TargetData info. Wrap it in 'if (TD)' in
preparation for the day we use null TargetData when no target is specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71210 91177308-0d34-0410-b5e6-96231b3b80d8

Optimize code placement in loop to eliminate unconditional branches or move unconditional branch to the outside of the loop. e.g.

///       A:
///       ...
///       <fallthrough to B>
///
///       B:  --> loop header
///       ...
///       jcc <cond> C, [exit]
///
///       C:
///       ...
///       jmp B
///
/// ==>
///
///       A:
///       ...
///       jmp B
///
///       C:  --> new loop header
///       ...
///       <fallthough to B>
///
///       B:
///       ...
///       jcc <cond> C, [exit]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71209 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing #include for "strlen" which is used inline in this header. Fixes
build under gcc 4.3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71208 91177308-0d34-0410-b5e6-96231b3b80d8

Moved pic16 naming functions to correct place.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71207 91177308-0d34-0410-b5e6-96231b3b80d8

PR4123: don't crash when inlining a call which uses its own result.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71199 91177308-0d34-0410-b5e6-96231b3b80d8

Fix pr4100.  Do not remove no-op copies when they are dead.  The register
scavenger gets confused about register liveness if it doesn't see them.
I'm not thrilled with this solution, but it only comes up when there are dead
copies in the code, which is something that hopefully doesn't happen much.

Here is what happens in pr4100: As shown in the following excerpt from the
debug output of llc, the source of a move gets reloaded from the stack,
inserting a new load instruction before the move.  Since that source operand
is a kill, the physical register is free to be reused for the destination
of the move.  The move ends up being a no-op, copying R3 to R3, so it is
deleted.  But, it leaves behind the load to reload %reg1028 into R3, and
that load is not updated to show that it's destination operand (R3) is dead.
The scavenger gets confused by that load because it thinks that R3 is live.

Starting RegAlloc of: %reg1025<def,dead> = MOVr %reg1028<kill>, 14, %reg0, %reg0
  Regs have values:
  Reloading %reg1028 into R3
  Last use of R3[%reg1028], removing it from live set
  Assigning R3 to %reg1025
  Register R3 [%reg1025] is never used, removing it from live set

Alternative solutions might be either marking the load as dead, or zapping
the load along with the no-op copy.  I couldn't see an easy way to do
either of those, though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71196 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment (again).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71180 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71179 91177308-0d34-0410-b5e6-96231b3b80d8

Revert 71165. It did more than just revert 71158 and it introduced
several regressions. The problem due to 71158 is now fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71176 91177308-0d34-0410-b5e6-96231b3b80d8

Perform constant folding on operands of instructions with non-void
types, such as loads and calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71175 91177308-0d34-0410-b5e6-96231b3b80d8

SCEVComplexityCompare's new code was missing SCEVUDivExpr. Implement
the SCEVUDivExpr case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71173 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r70876 and add a testcase (@c7) showing the problem:
bits captured, but the pointer marked nocapture.  In fact
I now recall that this problem is why only readnone functions
returning void were considered before!  However keep a small
fix that was also in r70876: a readnone function returning
void can result in bits being captured if it unwinds, so
test for this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71168 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert r71158. It was causing a failure during a full bootstrap:

checking for bcopy... no
checking for getc_unlocked... Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decUtility.c:360: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decUtility.o] Error 1
make[4]: *** Waiting for unfinished jobs....
Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decNumber.c:5591: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decNumber.o] Error 1
make[3]: *** [all-stage2-libdecnumber] Error 2
make[3]: *** Waiting for unfinished jobs....

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71165 91177308-0d34-0410-b5e6-96231b3b80d8

Make ScalarEvolution's GroupByComplexity more thorough. In addition
to sorting SCEVs by their kind, sort SCEVs of the same kind according
to their operands. This helps avoid things like (a+b) being a distinct
expression from (b+a).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71160 91177308-0d34-0410-b5e6-96231b3b80d8

Trim unnecessary headers. Code in Analysis shouldn't use Transforms
headers due to library dependencies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71159 91177308-0d34-0410-b5e6-96231b3b80d8

Constant-fold ptrtoint+add+inttoptr to gep when the pointer is an
array and the add is within range. This helps simplify expressions
expanded by ScalarEvolutionExpander.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71158 91177308-0d34-0410-b5e6-96231b3b80d8

Factor out a common base class between SCEVCommutativeExpr and
SCEVAddRecExpr. This eliminates redundant code for visiting
all the operands of an expression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71157 91177308-0d34-0410-b5e6-96231b3b80d8

Move the tablegen-produced DebugLoc handling into a AsmWriter::processDebugLoc function.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71156 91177308-0d34-0410-b5e6-96231b3b80d8

Code refactoring.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71151 91177308-0d34-0410-b5e6-96231b3b80d8

Rename "loop aligner" pass to "code placement optimization" pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71150 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate compiler warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71149 91177308-0d34-0410-b5e6-96231b3b80d8

Add const modifiers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71147 91177308-0d34-0410-b5e6-96231b3b80d8

THis doesn't fail.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71142 91177308-0d34-0410-b5e6-96231b3b80d8

Just turn aggressive stack coloring off at -O3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71140 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily revert r71010. It was causing massive failures during self-hosting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71138 91177308-0d34-0410-b5e6-96231b3b80d8

Make DwarfWriter::RecordInlinedFnStart more like the other DwarfWriter's methods:
-Have it return a label ID
-Remove the unused Instruction parameter

No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71132 91177308-0d34-0410-b5e6-96231b3b80d8

Use stable_sort instead of plain sort to avoid the risk of generating
trivially different code on different hosts (due to differing
std::sort implementations).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71124 91177308-0d34-0410-b5e6-96231b3b80d8

- Move some debug fields to coincide with how GCC emits them. No functionality
  change.
- Reformatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71118 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Use pthread library when requested and available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71116 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Updated lib/Target/PIC16/CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71115 91177308-0d34-0410-b5e6-96231b3b80d8

Use X86AddrNumOperands instead of magic constant one
more place.  This fixes a bunch of x86-64 JIT regressions.
(Introduced when the value of the magic constant changed
in 68645.  At the time apparently nobody noticed; failures
were hidden in 70343-70439 by an unrelated bug, so showed
up again as "new" failures in 70440.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71106 91177308-0d34-0410-b5e6-96231b3b80d8

Do not use register as base ptr of pre- and post- inc/dec load / store nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71098 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71091 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure to use signed arithmetic in APInt to fix a regression.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71090 91177308-0d34-0410-b5e6-96231b3b80d8

Simplify code by using SmallVector's pop_back_val() instead of
separate back() and pop_back() calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71089 91177308-0d34-0410-b5e6-96231b3b80d8

Add simplify_type specializations to allow WeakVH, AssertingVH, and
CallbackVH to participate in dyn_cast, isa, etc. without needing
an explicit conversion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71087 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Updated lib/CodeGen/CMakeLists.txt.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71085 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Detects libpthread and sets HAVE_LIBPTHREAD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71084 91177308-0d34-0410-b5e6-96231b3b80d8

CMake: Added cmakedefine for HAVE_PTHREAD_H.

Patch by Robert Schuster!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71083 91177308-0d34-0410-b5e6-96231b3b80d8

Nounwind is not valid for function return values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71082 91177308-0d34-0410-b5e6-96231b3b80d8

OCaml parameter attribute bindings from PR2752.
Incomplete, but better than nothing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71081 91177308-0d34-0410-b5e6-96231b3b80d8

Add generic expansion of SUB when ADD and XOR
are legal.  Based on a patch by Micah Villmow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71078 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR3754: don't mark functions that wrap MallocInst with
the readnone.  Since MallocInst is scheduled for deletion
it doesn't seem worth doing anything more subtle, such as
having mayWriteToMemory return true for MallocInst.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71077 91177308-0d34-0410-b5e6-96231b3b80d8

Emit banksel and movlp instructions.
Split large global data (both initialized and un-initialized) into multiple sections of <= 80 bytes.
Provide routines to manage PIC16 ABI naming conventions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71073 91177308-0d34-0410-b5e6-96231b3b80d8

Allow readonly functions to unwind exceptions.  Teach
the optimizers about this.  For example, a readonly
function with no uses cannot be removed unless it is
also marked nounwind.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71071 91177308-0d34-0410-b5e6-96231b3b80d8

A better error message.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71068 91177308-0d34-0410-b5e6-96231b3b80d8

Renamed Spiller classes (plus uses and related files) to VirtRegRewriter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71057 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate documentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71055 91177308-0d34-0410-b5e6-96231b3b80d8

The 'forward_as' property did not use its second argument.

See PR4159 for details. Patch by Martin Nowack!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71054 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a copy+pasto in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71035 91177308-0d34-0410-b5e6-96231b3b80d8

Delete a FIXME which is no longer relevant, and add a FIXME that is.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71033 91177308-0d34-0410-b5e6-96231b3b80d8

Quotes should be printed before private prefix; some code clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71032 91177308-0d34-0410-b5e6-96231b3b80d8

Add dump method to DIDescriptor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71028 91177308-0d34-0410-b5e6-96231b3b80d8

Add an explicit keyword.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71022 91177308-0d34-0410-b5e6-96231b3b80d8

If a MachineBasicBlock has multiple ways of reaching another block,
allow it to have multiple CFG edges to that block. This is needed
to allow MachineBasicBlock::isOnlyReachableByFallthrough to work
correctly. This fixes PR4126.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71018 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot this in the last commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71014 91177308-0d34-0410-b5e6-96231b3b80d8

Temporarily reverting r71008. It was causing this failure:

Running /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.src/test/
CodeGen/X86/dg.exp ...
FAIL: /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvm.src/test/
CodeGen/X86/change-compare-stride-1.ll
Failed with exit(1) at line 2
while running: grep {cmpq       $-478,} change-compare-stride-1.ll.tmp
child process exited abnormally

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71013 91177308-0d34-0410-b5e6-96231b3b80d8

Add some more documentation for x86 special address spaces.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71012 91177308-0d34-0410-b5e6-96231b3b80d8

Enable stack coloring with regs at -O3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71010 91177308-0d34-0410-b5e6-96231b3b80d8

Handle overflow of 64-bit loop conditions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@71008 91177308-0d34-0410-b5e6-96231b3b80d8

Add basic support for code generation of
addrspace(257) -> FS relative on x86.  Patch by Zoltan Varga!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70992 91177308-0d34-0410-b5e6-96231b3b80d8

bugpoint for jit should just ignore GCC arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70988 91177308-0d34-0410-b5e6-96231b3b80d8

Revert part of 70929 that has to do with determining whether a SIB byte is needed. It causes a lot of x86_64 JIT failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70986 91177308-0d34-0410-b5e6-96231b3b80d8